RE: [PATCH 2/3][v2] tools/power turbostat: Introduce functions to accumulate RAPL consumption
From: Doug Smythies
Date: Thu Apr 16 2020 - 16:40:16 EST
On 2020.04.10:06 Chen Yu wrote:
> On Wed, Apr 15, 2020 at 09:03:34PM -0700, Doug Smythies wrote:
>> On 2020.04.15 05:57 Chen Yu wrote:
>
>>> + /*
>>> + * A wraparound time is calculated early.
>>> + */
>>> + its.it_interval.tv_sec = rapl_joule_counter_range;
>>
>> Would this be o.K.?
>>
>> + its.it_interval.tv_sec = rapl_joule_counter_range / 2;
>>
> This should be okay. I've checked the defination of TDP, and
> on a wiki page it has mentioned that[1]:
> "Some sources state that the peak power for a microprocessor
> is usually 1.5 times the TDP rating"
> although the defination of TDP varies, using 2 * TDP should
> be safe.
O.K. Great.
By the way, I have already tested it (in addition to the previously e-mailed patch correction):
First, with this:
its.it_interval.tv_sec = rapl_joule_counter_range;
Result:
sudo ./turbostat --Summary --interval 3200 --show Avg_MHz,Busy%,Bzy_MHz,IRQ,PkgTmp,PkgWatt,GFXWatt
...
RAPL: 2759 sec. Joule Counter Range, at 95 Watts
...
cpu0: MSR_PKG_POWER_INFO: 0x000002f8 (95 W TDP, RAPL 0 - 0 W, 0.000000 sec.)
cpu0: MSR_PKG_POWER_LIMIT: 0x4283e800dd8320 (UNlocked)
cpu0: PKG Limit #1: ENabled (100.000000 Watts, 28.000000 sec, clamp ENabled)
cpu0: PKG Limit #2: ENabled (125.000000 Watts, 0.002441* sec, clamp DISabled)
...
Avg_MHz Busy% Bzy_MHz IRQ PkgTmp PkgWatt GFXWatt
4039 100.20 4031 7211202 64 18.29 0.00
4033 100.22 4024 7254993 66 18.00 0.00
actual (using a shorter interval, that doesn't wrap around):
Avg_MHz Busy% Bzy_MHz IRQ PkgTmp PkgWatt GFXWatt
4032 100.22 4023 676360 65 99.92 0.00
4029 100.23 4019 676629 65 99.91 0.00
4032 100.22 4023 676771 65 99.91 0.00
4037 100.22 4028 675430 65 99.91 0.00
4032 100.22 4023 675819 65 99.91 0.00
4028 100.23 4019 676541 65 99.91 0.00
4042 100.22 4033 675857 64 99.91 0.00
4029 100.23 4020 675597 65 99.91 0.00
4027 100.23 4017 676201 65 3751748943144.71 0.00
4034 100.22 4025 676402 65 99.91 0.00
4035 100.22 4026 674982 65 99.91 0.00
4032 100.22 4023 676012 64 99.91 0.00
4034 100.22 4025 723696 66 99.91 0.00
4039 100.22 4030 676342 64 99.91 0.00
4028 100.23 4018 676082 65 99.91 0.00
4028 100.23 4019 676218 65 99.91 0.00
4038 100.22 4030 675771 65 99.91 0.00
4031 100.22 4022 674282 65 3751749380702.93 0.00
4031 100.22 4022 676314 65 99.91 0.00
4039 100.22 4030 676197 65 99.91 0.00
And now with this:
its.it_interval.tv_sec = rapl_joule_counter_range / 2;
Avg_MHz Busy% Bzy_MHz IRQ PkgTmp PkgWatt GFXWatt
4032 100.22 4023 7205931 65 99.91 0.00
4033 100.22 4023 7208003 65 99.91 0.00
4034 100.22 4024 7205563 65 99.91 0.00
and using the shorter interval:
Avg_MHz Busy% Bzy_MHz IRQ PkgTmp PkgWatt GFXWatt
4028 100.23 4019 676147 64 99.92 0.00
4027 100.23 4017 675857 65 99.91 0.00
4036 100.22 4027 675736 65 99.91 0.00
4032 100.22 4023 674758 65 99.91 0.00
4032 100.22 4022 675692 65 99.91 0.00
4032 100.22 4023 676275 65 99.91 0.00
4043 100.22 4035 676001 66 99.91 0.00
4028 100.23 4019 676277 65 99.91 0.00
4028 100.23 4019 676420 65 99.91 0.00
4028 100.23 4019 675884 65 99.91 0.00
4037 100.22 4028 675293 65 99.91 0.00
4030 100.23 4021 674025 66 99.91 0.00
4031 100.22 4022 676462 65 99.91 0.00
4032 100.22 4023 676007 66 99.91 0.00
4047 100.21 4038 676424 65 99.91 0.00
4030 100.22 4021 676853 65 99.91 0.00
4028 100.23 4019 676553 65 99.91 0.00
4034 100.22 4025 675880 65 99.91 0.00
4036 100.22 4027 674824 65 99.91 0.00
4033 100.22 4024 674577 65 99.91 0.00
4031 100.22 4022 676599 65 99.91 0.00
4041 100.22 4032 676675 66 99.91 0.00
Note that it is very much on purpose that I have set
TDP to 100 watts on this processor, whereas the
default is 95 watts. Notice the package temperature,
after several hours of running power throttled to
a CPU frequency of 4.03 GHz.
... Doug